This is an author-deposited version published in: Eprints ID: 10622

Transcription

1 Open Archive Toulouse Archive Ouverte (OATAO OATAO is an open access repository that collects the work of Toulouse researchers and makes it freely available over the web where possible. This is an author-deposited version published : Eprts ID: 106 To cite this version: Msolli, Sabeur and Dalverny, Olivier and Alexis, Joël and Karama, Moussa Mechanical characterization of an Au-Ge solder alloy for high temperature electronic devices. (010 In: 6th International Conference on Integrated Power Electronics Systems (CIPS, 16 March March 010 (Nuremberg, Germany Any correspondence concerng this service should be sent to the repository admistrator: staff-oatao@listes-diff.p-toulouse.fr

2 Mechanical Characterization of an Au-Ge Solder Alloy for High Temperature Electronic Devices Sabeur Msolli, Université de Toulouse; INP/ENIT; LGP; Tarbes, France Olivier Dalverny, Université de Toulouse; INP/ENIT; LGP; Tarbes, France Joel Alexis, Université de Toulouse; INP/ENIT; LGP; Tarbes, France Moussa Karama, Université de Toulouse; INP/ENIT; LGP; Tarbes, France Abstract This paper presents a description of the mechanical behaviour of an Au-Ge solder under various loadg conditions as well as an elastoviscoplastic modellg. In order to achieve the modellg task, the solder is subjected to a set of shear, creep and fatigue loadgs order to determe the behaviour dependence to the temperature and displacement rate and which could be useful to evaluate the degradation of the material. These tests are realized under a wide range of temperatures, loads and displacement rates for modellg purposes. Then, a unified viscoplastic model is applied to predict correctly the material mechanical responses. The model is correlated with the experimental data 1 Introduction The need of more reliable power modules extreme temperature and high power environment is leadg to the vestigation of new range of solderg materials with more important thermomechanical and electrical characteristics. The choice of the solder is generally made with respect to the thermal and mechanical conditions as well as the chemical compatibility with underlyg components order to avoid the formation of voids or weak termetallic compounds the terfaces. There are many ways to ensure good bondg between module elements and the most terestg one is that which permits to reach the goal under low elaboration temperature. So, it s a great challenge to achieve a highly resistant jot without use of severe technical conditions and efforts. Among promisg techniques, paste silver sterg [1,] and solid state diffusion bondg (SSDB are now cited as substitutes for solders reflow process. Particularly, solid state diffusion bondg [3,4] is used for silver and gold based alloys such as Au-Sn and Au-Ge solders. It consists on a first reflowg step where the solder melts to form a primary jot between thick Au metalized components [5], then an annealg step to diffuse the bary entity on the metallization and mimize its proportion on the jot. Thus, we can obta a solid solution with much higher meltg temperature and good thermomechanical aspects. For this purpose, and as a first part of our extended work, we present next a mechanical characterization of an Au-Ge solder which may be used SSDB technique. The Au-Ge eutectic solder has good thermal properties and a meltg temperature of 356 C. Shear Behavior and modellg of the Au-Ge solder.1 Experimental procedure Cyclic displacement controlled tests are performed at 5 C, 00 C and 300 C under various displacement rates 10 -, 10-3 and 10-4 mm/s. Creep tests are also achieved at 5 C, 00 C and 300 C with imposed Load of 100 N and 00 N. The shear specimen considered consists of two soldered copper plates with a thickness of 3 mm as shown figure 1. The plates are soldered usg Au-Ge preform whose dimensions are 3 0. mm. The specimen was obtaed usg a reflow process of the Au-Ge solder alternated nitrogen-vacuum atmosphere to improve wettg quality and prevent oxidation. This specimen is then mounted on an Electroforce mache test as shown figure, which is rated up to 450 N and can perform precision materials tests cludg tension/compression loadg, fatigue and dynamic material characterization. Figure 1 Sgle shear jot specimen

3 Figure Test mache (left and specimen mountg (right The apparatus is equipped with a furnace which can reach temperatures up to 300 C. Displacements are measured by mean of an axial extensometer as shown figure.. Thermoviscoplastic modellg of the solder At extreme temperature i.e. 300 C, Au-Ge solder exhibit high stra rate sensitivity due to the activated creep mechanisms the materials. The solder is also subjected to thermal stressg due to the thermal expansions of the underlyg materials especially case of DBC copper plates. The teraction of plasticity and creep leads to the use of unified viscoplastic models which not separate the two phenomena and considers that they are issued from the same time-dependent deformation [6]. So, spite of the formulation complexity of these models, they can accurately reproduce a realistic behaviour of the material under complex loadg conditions. In this framework, a McDowell s unified thermoviscoplastic model [7] is used for the modellg task...1 Formulation of the model McDowell s viscoplastic model contas a set of constitutive equations: (1 a flow law which represents the material deformation and the viscosity effects and ( the evolution equations which describe the associated physical microstructural aspects usg a set of state variables. In this case, α represents the kematic hardeng variable and R is the isotropic hardeng variable. The flow law or the evolution of the elastic stra rate is expressed as follows 3 s α ε = AΘF ( ς, D, η (1 J σ α ( As we can see, the flow law depends on a diffusivity parameter and a Zener-Hollowman function F whose expression is fixed generally with the stress and temperature levels [8]. McDowell proposed an exponential term to take to account viscosity effects Where N 1 ( B N F = ς exp ς + ( J ς = ( σ α ( 1 η D R The factor η represents the sensibility partition of the stra rate and maly varies between 0 and 1. D is an isotropic resistance which is usually constant for solder alloys or proportional to the yield stress for copper ( D = A0 R. N is a temperature-dependent parameter. A and A 0 are temperature-dependent material parameters. The diffusivity parameter depends on the temperature level Q Tm Θ = exp for T kt (4 Q Tm T Θ = m exp ln + 1 for T ktm T T is the meltg temperature, Q is the activation energy, m and k is the Boltzmann s constant.. is defed as g = g if g 0 else g = 0. Microstructural changes such as dislocation motion, teraction between dislocations and isotropic defects repartition are described by two variables which are R and Kematic hardeng variable is divided to two other variablesα and α takg to account the displacement s * (3 rate of disclocations. In the isothermal case, we can write α ε α ε α (5 s s s s = Cb C CbΩ Θ * * * * * = H H Ω Θ α ε α (6 In the other hand, a variable constitute a couplg variable between isotropic and kematic hardeng effects i.e. the contribution of R iso and b iso followg iso 3 R = ( 1 ω (7 iso b = ω (8 ' = µ ( ε µω Θ (9 The variable saturates to and its evolution equation cludes dynamic and static thermal recovery terms such as those ofα and α s * C, H *, b 0, R 0, ' and are temperature-dependent parameters. The thermal static recovery terms are expressed the followg forms: M ( αα ( αα ( α α s / s s s s Ω = C, (10 * / * * * * M Ω = C, (11 M / Ω = C (1 The yield stress R and the kematic hardeng saturation iso b are written respectively as R = R 0 ( T + R and iso ( T b b = b 0 +

4 0 * s R is the itial yield stress. H, C, and M are material parameters. s * * M, C, M,.. Integration and implementation of the constitutive equations As for the most popular unified viscoplastic models, McDowell s model is formulated usg a set of strongly coupled, first order differential equations which is hard to solve without advanced tegration schemes. McDowell used a semi-implicit tegration scheme combed with a Newton-Raphson iteration procedure [9] to solve the model equations. In the semi-implicit scheme, the effective cumulated elastic stra crement p is updated implicit way contrarily to the stress and the other state variables. Moreover, an automatic steppg procedure is troduced order to improve the solution computation. Let s see that all the equations of the model may be written the form y = f ( y, t. So, the cremental solution of each variable is written as t+ dt t t t y = y + y = y + ty (13 t The aim is to mimize the error function F expressed as (,, ( t dt σ α ε : ( ε + t + dt F R = p t (14 t dt + Where( depends on the stress and state variables ε which are updated at t + t. The value ofp is updated for the next iteration k +1 only when the quantity ε tol k p p k+ 1 p k + 1 C is greater than a given tolerance fixed to 10-4 our case. The update of p is then computed usg the followg equation F 1 ( p, σ, α, R k + k p = p (15 F ' p,,, R ( σ α Otherwise, p is retaed for the next step as well as the correspondg stress and state variables for the iteration k+1. Algorithm developments as well as the tangent tensor computation are detailed elsewhere [9]. Once tegrated, the McDowell s model is implemented as subroute UMAT the fite element code ABAQUS. The solver used the fite element computation is an asymmetric one with a full Newton-Raphson iteration scheme...3 Material parameters identification An identification algorithm was performed usg Python scriptg based on the Least Square mimization procedure. This algorithm was combed to UMAT subroute to identify the model parameters with respect to the fite element model showed on the figure 3. Copper plates are considered as an elastoplastic material modeled with a combed hardeng elastoplastic Chaboche model. Elastic modulus and Poisson ratio of Au-Ge are taken respectively as E (MPa = T ( C and v = 0.3. For the fite element modelg, we consider the followg geometry figure 3 as the used model for the parameters identification and results extraction Figure 3 Fite element model for the identification procedure McDowell model contas only six temperaturedependent coefficients i.e. E, v, C, H *, b 0, R 0, ' and. All the temperature dependent parameters may be expressed as functions of temperature as for the Young s modulus E. The identified material parameters of the Au-Ge solder alloy are given the table 1 for McDowell s model McDowell s Model parameters Temperature 5 C 00 C 300 C n A B C D(MPa H*(MPa w C s C * (MPa/s M s M * Q(J/mol R 0 (MPa b 0 (MPa ' Table 1 Optimized model parameters for Au-Ge solder For the rate sensitivity viewpot, and followg experimental data shown figure 4 and 5, the material appears to be unaffected by displacement rate variation and shows no sensitivity to loadg rate at ambient temperature and 00 C. The material reaches all cases a load value of approximately 00N for a fixed displacement level at about 0.0 mm. Followg the results presented figures 6, 7 and 8, for the uniaxial shear tests, the soldered assembly shows a cyclic hardeng at least for the first cycle of loadg. Also, Bauschger effect is apparent which dicates kematic hardeng domation.

5 Figure 4 Uniaxial shear tests at various temperatures and stra rates Figure 7 Creep tests at 300 C and 100 N Figure 8 Cyclic shear tests at 5 C and 10-3 mm/s Figure 5 Creep tests at various temperatures and Loads Good correlation is also recorded between experimental data and numerical simulations obtaed by fite element computations especially for the cyclic shear tests which dicate acceptable set of identified model parameters. The modelg task may be extended usg thermomechanical fatigue and ratchetg tests to improve numerical and identification results. A damage model is also suitable for lifetime estimation of the solder material. 3 Conclusion A set of experimental tests are conducted to characterize the thermoviscoplastic behaviour of Au-Ge solder material. The data is the used to identify the McDowell s unified viscoplastic model which is successfully implemented Abaqus usg semi-implicit tegration algorithm. There was good agreement between numerical results from fite element simulations and tests results. That s dicates the efficiency of the identified model and thus it will be useful for more complex cases of loadg. 4 Literature [1] Bai, J.G. ; Y, J. ; Zhang, Z. ; Lu, G.Q. and Van Wyk, J.D. : High-Temperature Operation of SiC Power Devices by Low-Temperature Stered Silver Die- Attachment, IEEE Transcations On Advanced Packagg, Vol. 30, No. 3, August 007, pp Figure 6 Creep tests at 00 C and 100 N [] Bai, J.G. ; Calata, J.N. ; and Lu, G.Q. : Processg and Characterization of Nanosilver Pastes for Die-Attachg SiC Devices. IEEE Transcations On Electronics Packagg Manufacturg, Vol. 30, No. 4, October 007, pp

6 [3] Lee, C.C. ; Wang. C.Y.; and Matasevic, G. : Au-In Bondg Below the Eutectic Temperature. IEEE Transcations On Components, Hybrids, And Manufacturg Technology, Vol. 16, No. 3, May 1993, pp [4] Wang P.J., Kim J.S. and Lee C.C., A.: A New Bondg Technology Dealg with Large CTE Mismatch Between Large Si Chips and Cu Substrates. Electronic Components and Technology Conference 008, pp [5] Tjong, S.C. ; Ho, H.P. and Lee, S.T. : The terdiffusion of Sn from AuSn solder with the barrier metal deposited on diamond, Materials Research Bullet, Vol. 36, May 001, pp [6] Krausz, A. S. ; Krausz, K. : Unified constitutive laws of plastic deformation. Elsevier [7] McDowell, D.L.: A Nonlear kematic hardeng theory for cyclic thermoplasticity and cyclic thermoviscoplasticity, International Journal of Plasticity, Vol. 8, 199, pp [8] Stouffer, D.C. and Dame L.T. : Inelastic Deformation Of Metals: Models, Mechanical properties and metallurgy, John Wiley & Sons Inc, 13 février 1996 [9] Fu, C.; McDowell, D.L. and Ume, I.C. : A Fite Element Procedure of a Cyclic Thermoviscoplasticity Model for Solder and Copper Interconnects, Transactions of the ASME, Vol 10, 1998, pp